Computerized Medical Imaging and Graphics 37 (2013) 369–376 Contents lists available at ScienceDirect Computerized Medical Imaging and Graphics jo ur nal ho me pag e: www.elsevier.com/locate/compmedimag The use of radial symmetry to localize retinal landmarks A. Giachetti a,∗ , L. Ballerini b , E. Trucco b , P.J. Wilson c a Dipartimento di Informatica, Università di Verona, Italy b VAMPIRE, School of Computing, University of Dundee, UK c Department of Ophthalmology, Ninewells Hospital, Dundee, UK a r t i c l e i n f o Article history: Received 5 December 2012 Received in revised form 19 June 2013 Accepted 20 June 2013 Keywords: Optic disc Fovea Fast Radial Symmetry a b s t r a c t Locating the optic disc center and the fovea in digital fundus images is surprisingly difficult due to the variation range in color and contrast and the possible presence of pathologies creating bright spots or changing the appearance of retinal landmarks. These reasons make it difficult to find good templates of optic disc and fovea shape and color for pattern matching. In this paper we propose radial symmetry as the principal cue to locate both optic disc and macula cen- ters. Centers of bright and dark circularly symmetrical regions with arbitrary radii, can be found robustly against changes in brightness and contrast by using the Fast Radial Symmetry transform. Detectors based on this transform coupled with a weak hypothesis on vessel density (optic disc intersects large vessels while the fovea lies in an avascular region), can provide a fast location of both OD and macula with accuracy similar or better than state-of-the-art methods. The approach has been chosen as the default technique for fast localization of the two landmarks in the VAMPIRE software suite. © 2013 Elsevier Ltd. All rights reserved. 1. Introduction One of the most important tasks in retinal image processing is the location of the optic disc (OD) and of the macula, or, better, its central part, the fovea. The location of such landmarks is a prereq- uisite for the algorithms detecting signs of several retinal diseases. OD and fovea are often identified in digital fundus images with heuristics based on color and shape priors. The significant vari- ability of their appearance, however, makes it difficult to devise template based detectors working well on poor- or limited-quality images and in the presence of anomalies. Solutions proposed in the literature use heavily contextual information (e.g. vascularization). This makes the detector more robust, but can make the localization less accurate. Furthermore, anomalous cases may create problems in training multivariate models of objects and context. A possible alternative to a rigid modelling of the context is to find detectors that capture the peculiarities of the target landmarks better and are less sensitive to noise and disturbances, adding only a “soft” contextual reasoning to cope with anomalous cases. In this paper we show that radial symmetry is a simple and effec- tive cue to detect OD and fovea, and locate their centers. While it is difficult to find a good template for the fovea, and while differ- ent templates proposed so far for the quick detection of the OD do not outperform a simple bright circle, as shown by Yu et al. [1], we found that the use of a radial symmetry detector can considerably ∗ Corresponding author. Tel.: +39 045 8027998. E-mail address: andrea.giachetti@univr.it (A. Giachetti). improve robustness and accuracy with respect to other methods. Both optic OD and fovea have variable appearances (shape, con- trast with background, color components), but the OD, after vessel removal, is almost always characterized as a radially symmetrical bright spot, and the foveal region is better characterized as a radi- ally symmetric dark region than as the darker part of the image, or a part with a specific shape (e.g. a circle of specific radius). Fur- thermore, the OD should always be crossed by the largest vessels, while the center of the foveal region is avascular. We use the Fast Radial Symmetry (henceforth FRS) transform [2] to detect and localize centers of symmetry of dark and bright regions of arbitrary radius, independent of the presence of evi- dent contrast or edges. This allows us to design specific OD and fovea detectors, computing this transform on vessel-inpainted and coarsened images and combining the results with a vascular density estimator. This work has been carried on within the VAMPIRE project [3]. VAMPIRE (Vessel Assessment and Measurement Platform for Images of the REtina) is an international collaboration of 10 image processing and clinical centers, developing a software suite allowing efficient semi-automatic measurements on the retinal vasculature. 2. Related work OD and macula detection in digital fundus images is decep- tively simple: images quality can vary heavily in clinical samples of even modest size, lesions may create false targets, and even a normal OD is covered by vessels of irregular and variable shape. 0895-6111/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.compmedimag.2013.06.005